To your first point, if the price starts going against you significantly, the position is a bad one and your entry did not adequately trade the reversion. Again, the market action you are trying to capture is the issue here, i.e. in an un-hedged trade, donât buy a falling market, buy a reverting market. The distribution of residuals, while interesting in theory does not provide any better measure of recent value than the most recently available price. Also, the mean at time t+1 doesnât matter. Once a position is entered, position management becomes critical *irrelevant* of where the new mean may be. The reason is that now we are playing the expectancy game. The price might rocket up, or, it may go nowhere or down. Your entry, while likely a good low entry, does not forecast the outcome of the trade but should, over a large enough sample space, produce a positive expectancy outcome. Once the entry has taken place, your exit plan should be based on the probability of the price moving a certain distance for or against you. As you may have noticed this is not the pairs trading approach, it is a âbuy the reversionâ approach. âBasically, you're just gambling that E[S(t)] ~= E[S(t+1)]. â Well, I donât know about that⦠I think youâre getting stuck in the math of this rather than actual market behavior. Do you know what Impulse Response is? Think about what happens at sharp inflection points. What kind of price action do you usually see? Well, in the buy case, there is usually a quick move down (today's gap is a great example). The shorts have made money and where do you think the shorts put their stops? Just above recent highs if the move was sharp enough, right? IMO, trading mean reversion is trading where *profit* stops are being hit. Todayâs market action is a perfect example of what Iâm talking about. What happened at the open today and where did the price start to trade after the opening range? Itâs a concept thatâs tried and true⦠All that really mattered this morning was the gap and the fade. What type of entry wouldâve been ideal this morning? What Iâm trying to communicate is that arbitrarily trading forecasts where the current price X sigma away from some mean is not really a robust trading method. Additional logic is required. Iâve just outlined for you what a good mean reversion system consists of: 1. Identifying sharp moves. âToo quickâ in one direction or the other. 2. Identifying when the sharp move is over. A pause usually. 3. Identifying when the price starts reverting. In the buy case, price will slowly creep up, hitting tight stops early and then a vacuum starts. Ideally you want to trade those stops being hit. 4. Knowing the expectancy of your entry, i.e. the optimal sell point if youâre tricked by a false breakout.
If you want to see some really interesting heated debates about regression/inference in time series modeling, do a search on some scientist blogs for global warming and climate change. Their approaches generally put TA kindergarten math to shame, and they still are not in agreement about the "proper" way to model and predict a series.
I must admit that when I first started learning about the markets, I thought technical analysis was the way to go. I reasoned that even 99% of the population couldn't handle the basic math behind technical analysis. What I didn't count on was the fact that the markets are being traded by geniuses that have scaled their mathematics. It's the geniuses who're driving a lot of the valuation and making sure that, even if the EMH isn't true, that's it's approximately true within a relatively short period of time. Working in a derivatives group, I am absolutely in awe of how many years people went to school and the extremely complex mathematical models and modeling techniques they use. They consistently make money, year after year, while I see the retail trader bumbling around with patterns and shapes that have not been proven in any statistically satisfactory manner. Consider the simple moving average. You know that an MA(q) model assigns different weights to the the information in the past before calculating the average. There's also an infinite summation AR(p) form that's equivalent to the MA(q) form. Now, go over to a technical analysis package where the variable to be adjusted is simply the size of the window over which information is averaged in an unweighted fashion. TA approach for exponential moving averages also just assigns more urgency to recent points (better), but says nothing about any weight-adjustment schemes to account for independence between estimations of the means on a regression. I somewhat feel betrayed by technical analysis, and generally don't post in the TA forum. I just don't believe it works, outside of situational settings where it just so happens that an oscillator coincides with the points of a range where there happens to be a great deal of liquidity. ... Bah. end rant.
How does one know ahead of time if price will go against them when they enter? In other words, when price gets x distance from your mean, how do you know at that moment if it's a good time to enter or not? And similarly, how do you know when it comes time to place the trade if the market is reverting or not?